Qwen3.6 35B A3B qwen3.6-35b-a3b
qwen · qwen hybrid-reasoning efficient
#16 overall #12 agentic
- params
- 36B (A3.3B)
- arch
- moe
- context
- 256k
- license
- apache-2.0 open weights
- released
- Apr 2026
- reasoning
- yes
- downloads/30d
- 4.4M
# architecture
- layers
- 40
- d_model
- 2048
- heads
- 16q / 2kv
- head dim
- 256
- experts
- top-8 of 256
- vocab
- 248320
- family
- qwen3_5_moe_text
signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.
# why ranked
Overall: #16 — score 73.1 ● high — all signals present
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.70 | 73.7 |
| bench_composite | 0.30 | 71.6 |
benchmark panel evidence:
| complete panel | score (0–100) |
|---|---|
| aime-2026-v1 | 71.6 |
Coding: provisional — score 47.4 ● low — a single signal; treat with caution
provisional: too few independent signals for a numbered position — the score is shown, but this model sorts after every ranked model.
| signal | weight | input (0–100) |
|---|---|---|
| aa_coding | 0.30 | 47.4 |
| aider_polyglot | 0.30 | — |
| swe_bench_verified | 0.40 | — |
missing signals are dropped and the remaining weights renormalized — never imputed.
Agentic: #12 — score 50.0 ● low — a single signal; treat with caution
| signal | weight | input (0–100) |
|---|---|---|
| swe_bench_verified | 0.60 | — |
| swe_rebench | 0.40 | 50.0 |
missing signals are dropped and the remaining weights renormalized — never imputed.
# trend
methodology changed during this history window; score movement across that boundary is not model movement. See methodology v5.
# benchmarks
| benchmark | score | source | date |
|---|---|---|---|
| AA Coding Index | 41.9 | Artificial Analysis | — |
| AA Intelligence Index | 26.2 | Artificial Analysis | — |
| Ai2d | 0.9 / 1 | LLM Stats | — |
| AIME 2026 | 0.9 / 1 | LLM Stats | — |
| C-eval | 0.9 / 1 | LLM Stats | — |
| Cc-ocr | 0.8 / 1 | LLM Stats | — |
| Charxiv-r | 0.8 / 1 | LLM Stats | — |
| Claw-eval | 0.5 / 1 | LLM Stats | — |
| SWE-rebench (resolved) | 24.7 | SWE-rebench | — |
| SWE-rebench (pass@5) | 43.2 | SWE-rebench | — |
# where to run
| provider | quant | ctx | $/M in | $/M out | $/M cache | price src | tps | uptime | ✓ |
|---|---|---|---|---|---|---|---|---|---|
| Featherless | — | — | — | — | via hfrouter | — | — | ||
| Darkbloom | fp4 | 256k | $0.05 | $0.70 | — | via openrouter | — | 100.0% | |
| Libertai | 256k | $0.15 | $0.50 | — | via litellm | — | — | ||
| AkashML | fp8 | 256k | $0.10 | $0.90 | $0.05 | via openrouter | — | 100.0% | |
| DeepInfra | fp8 | 256k | $0.10 | $0.95 | — | deepinfra | — | 95.0% | ✓ |
| Venice | fp8 | 250k | $0.10 | $1.00 | — | venice | — | 100.0% | ✓ |
| Parasail | fp8 | 256k | $0.15 | $1.00 | $0.05 | via openrouter | — | 99.9% | |
| AtlasCloud | fp8 | 256k | $0.19 | $1.11 | — | atlascloud | — | 100.0% | ✓ |
| Io Net | fp8 | 256k | $0.19 | $1.19 | $0.09 | via openrouter | — | 100.0% | |
| Phala | unknown | 256k | $0.20 | $1.27 | — | phala | — | 98.7% | ✓ |
| CoreWeave | fp8 | 256k | $0.25 | $1.25 | $0.25 | via openrouter | — | 99.9% | |
| Wandb | 256k | $0.25 | $1.25 | — | via litellm | — | — | ||
| SiliconFlow | fp8 | 256k | $0.20 | $1.60 | — | openrouter | — | 99.2% | ✓ |
| Alibaba | 256k | $0.25 | $1.49 | — | via modelsdev | — | — | ||
| Novita | 256k | $0.25 | $1.49 | — | novita | — | — | ✓ | |
| Scaleway | — | $0.28 | $1.71 | — | via hfrouter | 155 | — |
sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price
source aliases
- aa
qwen3.6-35b-a3b- hf
Qwen/Qwen3.6-35B-A3B-FP8,nvidia/Qwen3.6-35B-A3B-NVFP4- openrouter
qwen/qwen3.6-35b-a3b